Haiku 4.5

Name: Haiku 4.5
Author: Anthropic

Anthropic · 2025-10-15

Haiku 4.5 is Anthropic's fast, cost-efficient small Claude model for low-latency and high-throughput use, supporting extended thinking and computer use. It approaches Sonnet 4-level coding ability at a fraction of the cost and roughly double the speed.

Benchmark scores

Benchmark	Score
A-Fantasia	66.7
A-Fantasia (Valid Attempts with Retries)	58.0
A-Fantasia Backwards Spelling	69.0
A-Fantasia Chess	52.0
A-Fantasia Chess (Valid Attempts with Retries)	41.0
A-Fantasia Cube (Valid Attempts with Retries)	68.0
A-Fantasia Cube Rotation	79.0
A-Fantasia Spelling (Valid Attempts with Retries)	65.0
AA-AnalystAgent (pass@1)	30.0
AA-AnalystAgent (pass@5)	51.3
AA-AnalystAgent (pass^5)	15.0
AA-Briefcase Elo	614
AA-Briefcase v1.1 Elo	616
AA-LCR	70.3
AA-LCR v1.1	74.3
AA-Omniscience Index	-4.4
Agent Red Teaming k=100 (CAIS)	69.7
Aikido Known-CVE Detection (Pass@3 Union)	53.8
Aikido Known-CVE Detection (Single-Run Average)	41.0
AIME 2024 (avg@8)	83.8
AIME 2024-2025	82.7
AIME 2025	80.7
AIME 2025 (Artificial Analysis avg@10)	83.7
AIME 2025 (avg@8)	81.7
AIME 2026	85.1
ALE-Bench	1006
Antidote: Everyday Edition	932
ARC-AGI-1 Semi-Private	47.7
ARC-AGI-2	4.0
Arena Code WebDev — Brand & Marketing	1323